实战篇:PostgreSQL开发与优化(4)查询性能提升
发表时间: 2024-03-04 13:47
作者:太阳
查询当前正在运行的会话中耗时最长的Top SQL,where条件可按需修改
SELECT pgsa.datname AS database_name , pgsa.usename AS user_name , pgsa.client_addr AS client_addr , pgsa.application_name AS application_name , pgsa.state AS state , pgsa.backend_start AS backend_start , pgsa.xact_start AS xact_start , extract(epoch FROM now() - pgsa.xact_start) AS xact_time, pgsa.query_start AS query_start , extract(epoch FROM now() - pgsa.query_start) AS query_time , pgsa.query AS query_sqlFROM pg_stat_activity pgsaWHERE pgsa.state != 'idle' AND pgsa.state != 'idle in transaction' AND pgsa.state != 'idle in transaction (aborted)'ORDER BY query_time DESCLIMIT 20;
pg_stat_activity视图各字段含义:
http://postgres.cn/docs/13/monitoring-stats.html#
MONITORING-PG-STAT-ACTIVITY-VIEW
pg_stat_statements可跟踪服务器执行的所有SQL语句的计划信息和执行统计信息,详见官方文档说明:[http://postgres.cn/docs/13/pgstatstatements.html](http://postgres.cn/docs/13/pgstatstatements.html)
1)修改配置文件
# su - postgres$ vi /data/pgsql13/data/postgresql.confshared_preload_libraries='pg_stat_statements,pg_pathman' pg_stat_statements.max = 10000pg_stat_statements.track = all2)重启生效:$ pg_ctl -D /data/pgsql13/data restart
3)载入pg_stat_statement插件
postgres=# \xExpanded display is on.--查看可用模块postgres=# SELECT * FROM pg_available_extensions WHERE name = 'pg_stat_statements';-[ RECORD 1 ]-----+-----------------------------------------------------------------------name | pg_stat_statementsdefault_version | 1.8installed_version | comment | track planning and execution statistics of all SQL statements executed--载入模块,载入后pg_stat_statements表可正常使用postgres=# create extension pg_stat_statements; CREATE EXTENSION
postgres=# select * from pg_stat_statements limit 1;-[ RECORD 1 ]-------+-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------userid | 16480 //用户oiddbid | 163959 //数据库oidqueryid | -7584655433466348220 //查询idquery | SELECT ...... //SQL语句plans | 0 //计划语句的次数,单位ms(启用pg_stat_statements.track_planning才会统计,默认关闭)total_plan_time | 0 //计划语句所花费的总时间,单位ms(启用pg_stat_statements.track_planning才会统计,默认关闭)min_plan_time | 0 //计划语句所花费的最短时间,单位ms(启用pg_stat_statements.track_planning才会统计,默认关闭)max_plan_time | 0 //计划语句所花费的最长时间,单位ms(启用pg_stat_statements.track_planning才会统计,默认关闭)mean_plan_time | 0 //计划语句所花费的平均时间,单位ms(启用pg_stat_statements.track_planning才会统计,默认关闭)stddev_plan_time | 0 //计划语句花费的时间的总体标准偏差,单位ms(启用pg_stat_statements.track_planning才会统计,默认关闭)calls | 92545 //语句被执行的次数total_exec_time | 1563.9699899999862 //执行语句所花费的总时间,单位msmin_exec_time | 0.005605 //执行语句所花费的最短时间,max_exec_time | 7.055763 //执行语句所花费的最长时间,mean_exec_time | 0.01689956226700567 //执行语句所花费的最长时间,stddev_exec_time | 0.036137014177393116 //执行语句花费的时间的总体标准偏差,rows | 17277 //语句检索或影响的总行数shared_blks_hit | 394706 //语句的共享块缓存命中总数shared_blks_read | 7 //语句读取的共享块总数shared_blks_dirtied | 6 //被语句弄脏的共享块总数shared_blks_written | 0 //语句写入的共享块总数local_blks_hit | 0 //语句的本地块缓存命中总数local_blks_read | 0 //语句读取的本地块总数local_blks_dirtied | 0 //被语句弄脏的本地块总数local_blks_written | 0 //语句写入的本地块总数temp_blks_read | 0 //语句读取的临时块总数temp_blks_written | 0 //语句写入的临时块总数blk_read_time | 0 //语句读取块所花费的总时间blk_write_time | 0 //语句写入块所花费的总时间wal_records | 7874 //语句生成的 WAL 记录总数wal_fpi | 5 //语句生成的 WAL 整页图像总数wal_bytes | 450177 //语句生成的 WAL 字节总数
oid是唯一标识,查询用户名与用户oid的关系:
postgres=# select userid,userid::regrole from pg_stat_statements group by userid;-[ RECORD 1 ]----userid | 10userid | postgres...
查询Top SQL:
-- 按总执行时间查询Top SQLselect userid::regrole as user_name,* from pg_stat_statements order by total_exec_time desc limit 20;-- 按总IO消耗查询Top SQLselect userid::regrole as user_name,* from pg_stat_statements order by blk_read_time+blk_write_time desc limit 20;-- 按总调用次数查询Top SQLselect userid::regrole as user_name,* from pg_stat_statements order by calls desc limit 20;
pg_stat_statements是累积的统计,累积的是pg_stat_statements配置后至当前时刻,无法查询这期间指定时间范围内的Top SQL情况;
可通过如下命令,清理历史统计信息:
select pg_stat_statements_reset();
通过定时清理历史统计信息+定时查询pg_stat_statements的方式可实现查询指定时间范围内的Top SQL;
如在每天0点清理历史统计信息,在每天9、11、17点分别查询pg_stat_statements,可得到每天0~9、0~11、0~17这3个时间范围内的Top SQL。
更多技术信息请查看云掣官网云掣YunChe - 可观测运维专家 | 大数据运维托管 | 云MSP服务